基于多教师知识蒸馏的变工况轴承故障诊断方法  

A Method for Bearing Fault Diagnosis under Variable Working Conditions Based on Multi-teacher Knowledge Distillation

作　　者：丁建建 朱晓娟[1] DING Jian-jian;ZHU Xiao-juan(School of Computer Science and Engineer,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《辽东学院学报（自然科学版）》2024年第1期31-39,共9页Journal of Eastern Liaoning University:Natural Science Edition

基　　金：安徽高校自然科学研究重点项目(KJ2020A0300)。

摘　　要：轴承在不同运行工况下所产生的故障数据存在分布差异,导致训练的智能故障诊断模型的诊断性能不佳。为此,提出一种基于多教师蒸馏的变工况轴承故障诊断方法。基于多源域迁移将所提方法分为2个阶段进行迁移:先将多个工况知识迁移到中间域模型中,聚合源域通用知识后再通过目标域中少量样本对模型进行微调,完成目标域适应;再使用集成分类器将得到的公共特征进行故障分类,得到故障诊断结果。实验结果表明,所提方法可在少量目标工况标记样本条件下对不同工况故障进行有效诊断,具有较高的准确性和泛化性。There are distributional differences in the fault data generated by bearings under different operating conditions,leading to poor diagnostic performance of the trained intelligent fault diagnosis model.A method for bearing fault diagnosis under variable working conditions based on multi-teacher knowledge distillation was proposed.This method was divided into two phases based on multi-source domain migration.Firstly,multiple working condition knowledge was migrated to the intermediate domain model,then general knowledge of the source domain was aggregated aggregated and a small number of samples in the target domain were used to fine-tune the model to complete the target domain adaptation,and finally the obtained common features were classified with an integrated classifier and the fault diagnosis was acquired.The experimental results show that the proposed method can achieve fault diagnosis under different working conditions with high accuracy with only a small number of target condition marked samples.

关 键 词：故障诊断 知识蒸馏 多源域迁移 变工况 

分 类 号：TH133.33[机械工程—机械制造及自动化]